Correlation vs Causation in Health Tracking

Your data shows a pattern: on days you drink coffee, you sleep worse. Coffee causes bad sleep, right?

Maybe. Or maybe you drink coffee on stressful days, and stress causes bad sleep. Or maybe you drink coffee when you slept poorly the night before.

Correlation isn't causation. Understanding the difference changes how you interpret your data.

The Basic Distinction

Correlation

Two things happen together. When A happens, B tends to happen too.

Examples:

• Ice cream sales and drowning deaths correlate (both increase in summer)
• Shoe size and reading ability correlate in children (both increase with age)
• Coffee consumption and poor sleep correlate

Correlation tells you things are related somehow. It doesn't tell you why.

Causation

A actually causes B. If you change A, B changes as a result.

Examples:

• Studying causes better test scores (if you study more, scores improve)
• Exercise causes fitness improvement (if you exercise, you get fitter)
• Sleep deprivation causes tiredness (less sleep, more tired)

Causation means intervention works: change the cause, change the effect.

Key Insight: Correlation is common. Causation is what you need for making changes.

Why This Matters for Health Tracking

You Want to Change Things

When you track health data, you want to find causes:

• What actually affects my weight?
• What truly improves my sleep?
• What genuinely changes my blood pressure?

If you find correlation but mistake it for causation, you'll change the wrong things.

False Confidence

Data makes us confident. "The numbers prove it!"

But numbers show correlation easily. Proving causation is much harder.

Wasted Effort

If coffee correlates with poor sleep but doesn't cause it, eliminating coffee won't help your sleep—but you'll have given up coffee for nothing.

Finding actual causes = effective changes.

Common Correlation Traps

Reverse Causation

You see: Exercise correlates with better mood You assume: Exercise causes better mood But maybe: Good mood causes exercise (you exercise when you feel good)

The relationship exists, but the direction might be backward.

Confounding Variables

You see: Reading before bed correlates with better sleep You assume: Reading causes better sleep But maybe: Avoiding screens (to read) causes better sleep, and reading is just what you do instead

Something else might be the real cause.

Coincidence

You see: You ate fish twice last week and felt great both times You assume: Fish improves how you feel But maybe: Random chance; two data points mean nothing

Small samples produce spurious correlations.

Selection Bias

You see: Days you track carefully correlate with better outcomes You assume: Detailed tracking causes better outcomes But maybe: You only track carefully on good days, when you have energy

What you measure when affects what you find.

Real Examples in Health Tracking

Weight and Sleep

Correlation: Poor sleep correlates with higher weight

Possible explanations:

1. Poor sleep causes weight gain (hormonal effects)
2. Higher weight causes poor sleep (sleep apnea)
3. Stress causes both poor sleep and weight gain
4. Busy schedules cause both (less sleep, more convenience food)

All four explanations fit the data. Only experiments can determine which is true for you.

Coffee and Energy

Correlation: Coffee correlates with more energy

Possible explanations:

1. Coffee causes energy (caffeine)
2. You drink coffee when you need energy (low energy causes coffee)
3. Morning routine includes coffee and naturally higher energy
4. Placebo effect

The first explanation is probably right (we know caffeine works), but the relationship is complex.

Exercise and Blood Pressure

Correlation: Regular exercise correlates with lower blood pressure

Good news: This one is well-established as causal. Exercise genuinely lowers BP.

But: The correlation in your data might also reflect: people who exercise tend to eat better, sleep better, and stress less. All of those affect BP too.

How to Think About Your Data

Start with Correlation

Correlation is the first step. You notice patterns:

• When X happens, Y tends to happen
• Weeks with more A have less B
• Input Z correlates with better outcomes

These are hypotheses, not conclusions.

Consider Alternative Explanations

For every correlation, ask:

• Could the direction be reversed?
• Is there a confounding variable?
• Could this be coincidence?
• Is my sample large enough?

The goal isn't skepticism for its own sake. It's avoiding false conclusions.

Test Through Experiments

The best way to find causation: N-of-1 experiments.

Change one variable:

• Week 1: Normal (no change)
• Week 2: Remove suspected cause
• Week 3: Back to normal
• Week 4: Remove again

If the outcome changes consistently when you change the input, causation becomes more likely.

Use Domain Knowledge

Not all correlations are equal. Consider:

• Does this make biological sense?
• Is there research supporting this mechanism?
• Have others found this causal relationship?

Coffee causing alertness is well-established. Wearing blue causing weight loss isn't.

Practical Guidelines

Strong Evidence for Causation

When your data probably shows causation:

• Large sample (many days/weeks)
• Consistent pattern
• Plausible mechanism
• Matches established research
• Experimental confirmation

Weak Evidence for Causation

Probably just correlation:

• Small sample (few observations)
• Inconsistent pattern
• No plausible mechanism
• Contradicts established research
• Confounders obviously present

The In-Between

Most of your data will be uncertain:

• Reasonable sample size
• Somewhat consistent pattern
• Plausible but unconfirmed mechanism

This is where experiments help.

Making Decisions Under Uncertainty

Low-Cost Changes

If the intervention is easy and low-risk:

• Try it even if causation is uncertain
• The cost of being wrong is low
• Example: Trying to sleep earlier when late nights correlate with fatigue

High-Cost Changes

If the intervention is hard or risky:

• Require stronger evidence
• Consider experiments first
• Example: Major dietary changes when certain foods correlate with problems

Reversible vs. Irreversible

Reversible: If it doesn't work, you can change back

• Lower evidence threshold acceptable
• Example: Trying different bedtimes

Irreversible: You can't undo it

• Require strong evidence
• Example: Medical decisions (always involve your doctor)

Common Health Correlations

Probably Causal

Well-established relationships:

• Sleep duration and energy (more sleep = more energy)
• Exercise and fitness (exercise = better fitness)
• Sodium and blood pressure (more sodium = higher BP)

You can act on these with reasonable confidence.

Mixed Evidence

More complex relationships:

• Meal timing and weight (correlation clear, causation debated)
• Stress and most outcomes (relationship real, mechanisms complex)
• Weather and mood (correlation exists, individual variation high)

Worth tracking and experimenting.

Often Spurious

Correlations that often don't hold up:

• Specific foods and dramatic effects (usually confounded)
• Moon phases and anything (coincidence)
• Minor supplement variations and outcomes (noise)

Be skeptical until proven otherwise.

Your Data's Limitations

Sample Size

Even a year of tracking is ~365 data points. That's decent for strong effects, but subtle causations need more data or careful experiments.

Confounders Everywhere

Real life is complex. When you sleep poorly, you also:

• Make different food choices
• Exercise less
• Feel more stressed
• Behave differently in dozens of ways

Isolating variables is hard.

Measurement Error

Your data isn't perfect:

• Weight fluctuates
• Sleep tracking isn't precise
• Subjective ratings vary
• You forget things

This noise can create false correlations or hide real ones.

The Humble Approach

Hold Conclusions Loosely

Your data suggests patterns. Those patterns might be causal.

But stay humble:

• New data might contradict
• Experiments might disprove
• You might have missed something

Iterate

Science is iterative:

1. Notice correlation
2. Form hypothesis
3. Test hypothesis
4. Revise understanding
5. Repeat

Your personal health tracking is your own science.

Action Despite Uncertainty

You can't wait for certainty. Perfect information never comes.

Make reasonable bets:

• Likely causes (act on them)
• Uncertain causes (experiment)
• Unlikely causes (be skeptical)

Next Steps

• Read: N-of-1 Experiments: You Are Your Own Control Group
• Read: Understanding Correlations in Trendwell
• Review: Your current correlations with skeptical eye
• Design: An experiment to test a suspected cause
• Remember: Correlation is the start, not the answer

Your data shows patterns. The question is always: what do they mean?

---

Last updated: January 2026